More on efficiency and genAI

I sometimes grit my teeth and check-in with genAI to see how its doing, and today was such a day. I had to write a letter of support for a funding application by a prospective postgraduate researcher.

Normal workflow

Open folder and find most recent similar letter. Make a copy, change names, funders etc., check details about myself and facilities we offer are uptodate, write a paragraph about project, add a few sentences about applicant, convert to PDF. Time = 30-40 minutes (depending on whether I have to refresh my memory about the project).

genAI workflow

Write prompt including pasted details of applicant, project and funder requirements. Read first output and ask for changes. Copy second output into letter template. Tweak formatting. Change phrasing here and there where it wasn't quite right. Convert to PDF. Time = 15 minutes. ¹

So that saved time?

Yes it did. Or more precisely it reduced wasted time. To be completely honest, the process of writing a letter of support like this is pointless since the relevant information it conveys to the funder (I am aware of applicant and project, happy to supervise, and can provide the most basic facilities any good university can provide) could equally easily be conveyed by an online form with four or five checkboxes which would take less than five minutes to complete. Headed paper, titles, positive, even superlative, adjectives etc. are all meaningless theatre every candidate for funding will have. The fact I sent the letter to the applicant for them to submit means the funder doesn't even know it came from me. A verified institutional email address and a quick online form would have been a better solution for them, let alone me.

Reflections

I was impressed by the output (it was Gemini 2.5 Pro, if you are curious) - it read like any of the hundreds and hundreds of these I have seen or written in my professional life. In fact, at one point I wondered if one or more of my own such letters had got into the training data, it was so similar to what I normally write. But then, so are letters written by other people.

This just highlights that the process was largely pointless. We don't need AI to get efficiencies here. In fact, improved processes would yield greater efficiencies. However, what the genAI approach to efficiency does is shift costs around. The funder has an inefficient process but it would cost them money to improve it. So they push most of the inefficiency out onto the people who benefit from their funding. genAI then allows those people to push most of that inefficiency away from their time and onto planetary resources (plus the Software-as-a-Service licensing costs), via an economic model which allows technology firms to not pay for externalities in the name of economic growth.

What happens is that the inefficiency of the system continues, but is masked, is made to feel less frustrating to the individuals having to implement it, thereby reducing the pressures to change the system in order to make it more efficient. But something has changed: The system was inefficient by being labour-intensive and when some of that labour is replaced by genAI, it becomes more capital-intensive, but equally inefficient system. That change benefits the owners of capital and harms the owners of labour. Inefficient systems which create more jobs serve a positive social function,² but inefficient systems which merely create more wealth for the already wealthy owners of capital appear to do social harm, reducing the living standards of the poorer and allowing the wealthy to distort the democratic process.

Of course, there are powerful forces at work which are very good at distracting us from this larger view. We are all busy and under pressure and grateful for everything that makes our lives a bit easier in the here and now. Each individual is tempted to use genAI to make their own lives a little bit easier under the illusion that these efficiency savings accumulate and make the whole system more efficient. But they don't. Yet no one individual or organisation is motivated to improve the efficiency of the whole system. We have a classic collective action problem.